Genetic heterogeneity is one of the major reasons for failure to identify genetic associations of complex diseases. Often, patients with complex diseases have various phenotypic characteristics and can be grouped into variable subtypes, possibly reflecting differences in underlying genetic mechanisms. Existing approaches either ignore genetic heterogeneity among patients, or lack parsimony with a large number of degrees of freedom in test statistics. There is a lack of statistical approaches that can efficiently combie association evidence from varied disease subtypes while accounting for genetic heterogeneity. As such, we propose to test genetic association using a novel three-stage polynomial logistic regression model, which takes genetic heterogeneity among disease subtypes into account while reducing large number of parameters for testing genetic association. We plan to apply the proposed approach to a real dataset from a collaboration study with the goal to find genetic associations of structural cardiovascular malformations in 22q11DS children. We expect that the proposed project will yield a new powerful statistical approach and the corresponding software for identifying genetic associations of complex diseases, and has the potential to identify novel genetic variants, genes and pathways, providing an insight into biological mechanisms of congenital heart defects.